A modular AI agent framework built with a TypeScript monorepo and a Python AI runtime. Designed for provider abstraction, streaming, tool calling, and extensibility, making it easy to integrate new models, tools, and agent capabilities.

Overview

Mini Agent is a modular AI agent framework designed to explore how production-grade AI assistants are built.

The project separates orchestration, language model providers, tools, and shared infrastructure into independent packages, making it easy to extend the system with new capabilities while maintaining a clean architecture.

Although still under active development, the framework already supports configurable providers, streaming responses, persistent configuration, and a developer-friendly CLI experience.

Architecture

The project follows a monorepo architecture with clear separation of responsibilities.

CLI – Interactive command-line interface for chatting with and configuring the agent.
Agents – Conversation management, prompt construction, and agent orchestration.
Providers – Pluggable LLM providers with a common interface.
Core – Dependency management, configuration, and application wiring.
Tools – Extensible tool registry for future agent capabilities.
Shared – Common types, models, and configuration shared across packages.
Python AI Runtime – Dedicated backend responsible for model inference and AI-specific logic.

Current Features

Modular monorepo architecture
Interactive CLI interface
Streaming and full-response modes
Provider abstraction
Easy provider switching
Configurable models and temperature
Persistent user configuration
Conversation management
Prompt builder abstraction
Python AI runtime integration
Tool registry foundation
Type-safe shared interfaces

Why I Built This

Most AI applications tightly couple the user interface, business logic, and language model provider.

Mini Agent takes a different approach by separating responsibilities into reusable modules, allowing providers, tools, and future capabilities to evolve independently.

The long-term goal is to build an extensible agent platform rather than a single-purpose chatbot.

Roadmap

✅ Implemented

Monorepo architecture
Modular package system
CLI interface
Provider abstraction
Streaming support
Python inference service
Configuration management
Conversation history
Tool registry foundation

🚧 Currently Building

Advanced tool calling
Memory system
Planning engine
Multi-step reasoning
Agent workflows
Browser automation
File system tools
Code execution
Additional providers

📅 Planned

Multi-agent collaboration
Long-term memory
Retrieval-Augmented Generation (RAG)
MCP support
REST API
Telegram integration
VS Code integration
Web dashboard

Key Takeaway

Mini Agent is my flagship AI engineering project and an ongoing exploration into building production-ready AI agents from first principles. Every iteration focuses on improving modularity, extensibility, and developer experience while gradually incorporating more advanced agent capabilities.